|M.Sc Student||Alul Orit|
|Subject||Multi-Dimensional Reputation in Open Multi-Agent Systems|
|Department||Department of Industrial Engineering and Management||Supervisor||Dr. Onn Shehory|
The proliferation of e-commerce systems on the web has laid the ground for the introduction of e-commerce agents that trade on behalf of their users. In many trade scenarios, parties need to collaborate and cooperate within different environments. In order to achieve such collaboration between two parties (and in particular among software agents), which possibly have never encountered each other in the past some trust has to be established. A common way to enhance trust among interacting parties on the web is by the use of reputation systems. Such systems are also used by known e-commerce sites such as eBay and Amazon. However, present reputation systems are vulnerable to several simple attacks. In this study we expose the shortcomings of existing reputation mechanisms by presenting an array of attacks to which they are vulnerable. Then, we design a new reputation management system that should alleviate such attacks and in some cases, even eliminate them. We examine the properties of this system and evaluate it by performing simulation experiments in comparison to previous solutions.
According to the outcomes of the simulation experiments we have run, we conclude that our reputation management system allows buyers to distinguish between trustworthy and non-trustworthy sellers, and in addition encourages sellers to be trustworthy. Furthermore, the suggested reputation management system alleviates the existence of attacks. As our experiment show, agents that use the suggested reputation management system arrive at gains greater than those arrived at by using alternatives such as eBay-like reputation management system or random choosing seller.